Social Identity and Inequality:
The Impact of China’s Hukou System
Farzana Afridi
Planning Unit
Indian Statistical Institute
7, S.J.S. Sansanwal Marg
New Delhi, 110016, India
fafridi@isid.ac.in
Sherry Xin Li
(Corresponding author)
School of Economic, Political and Policy Sciences (EPPS)
University of Texas at Dallas, GR31
800 W. Campbell Rd.
Richardson, TX 75080
Phone: 972-883-4987
Fax: 972-883-6486
sherry.xin.li@utdallas.edu
Yufei Ren
Department of Economics
Union College
807 Union St.
Schenectady, NY 12308
reny@uion.edu
Social Identity and Inequality:
The Impact of China’s Hukou System
Abstract
We conduct an experimental study to investigate the causal impact of social identity on individuals’
performance under incentives. We focus on China’s household registration (hukou) system, which
favors urban residents and discriminates against rural residents in resource allocation. Our results
show that making individuals’ hukou identity salient significantly reduces the performance of rural
migrant students, relative to their local urban counterparts, on an incentivized cognitive task, and
consequently significantly lowers their relative ranking in the earnings distribution under the piece
rate regime. However, the impact of hukou identity salience is insignificant in the tournament
regime, suggesting that its negative effect on migrant students’ performance may be mitigated
when competition is introduced. The results demonstrate the impact of institutionally imposed
social identity on individuals’ economic performance, and potentially on inequality.
Keywords: social identity, inequality, field experiment, hukou, China
JEL Classification: C93, D03, O15, P36
1
1.
Introduction
A large body of literature documents significant and increasing economic inequality in the
emerging economies of the world (in China: Kanbur and Zhang, 1999; Chen, 2002; Yao, Zhang
and Hanmer, 2004; Ravallion and Chen, 2007; in India: Deaton and Dreze, 2002; Datt and
Ravallion, 2002). Akerlof and Kranton (2000) incorporate individuals’ social identity into a
theoretical model of poverty and show that social exclusion can lead to equilibria in which the
‘excluded’ individuals avoid economic activities that are remunerative. Thus, if we incorporate the
psychology and sociology of an individual’s identity into economic models of behavior the
implications for economic inequality may be much wider.
This paper extends the literature on social identity by investigating whether individuals’
identification with an institutionally created underclass can contribute to widening the income gap
between them and the privileged. We focus on China’s household registration system – the hukou
institution. This system categorizes citizens into urban (non-agricultural) and rural (agricultural)
residents of a particular location. The urban residents in that location (say a municipality) are
favored in resource allocation compared to the rural residents and migrants. We design a framed
field experiment (Harrison and List, 2004) to study whether individuals’ identities created by this
institution affect their performance under incentives, and the distribution of earnings among these
different socio-economic groups.
To introduce an exogenous variation in identity salience we adopt a methodology from
psychology called priming (Bargh, 2006). Specifically, we randomly assign primary school
students in Beijing, with different hukou backgrounds, to two treatments. In the identity salience
treatment we prime students’ hukou identity and make it salient through a pre-experiment
questionnaire followed by a public verification of their hukou status (Shih, Pittinsky and Ambady,
1999; Hoff and Pandey, 2006, 2012). In the control treatment students’ hukou identity is kept
2
private. We then investigate the causal impact of hukou identity salience by comparing the
difference in the performance of local urban students and students from rural migrant families in
incentivized puzzle-solving tasks between the two treatments, i.e., by using a difference-indifference approach. We also examine how the impact of hukou salience on the performance of the
two groups translates into widening the experimental earnings gap between them.
Our experimental design follows Hoff and Pandey (2006, 2012) who were the first to explore
how social identity contributes to inequality through its impact on individuals’ performance on
incentivized tasks in India. This paper extends their investigation to China – the world’s most
populous country and fastest growing economy – and more importantly, to a broader horizon in
several ways. First, we explore the impact of identity within a socio-political system that is almost
polar opposite of India’s. Unlike India, China is an ethnically homogeneous society – its population
is 91.5 percent ethnic Han (National Bureau of Statistics 1994-2003). While the caste system is
based on factors that include ethnicity and occupation, the rural-versus-urban categorization by
China’s hukou system is less complex. The relatively simple social categorization by hukou,
therefore, provides cleaner evidence on how powerful institutional exclusion can be, even in an
ethnically homogeneous society, in influencing labor market outcomes through individuals’ social
identities. Second, like the caste system, hukou is an ascriptive order (i.e. it is assigned by birth)
but unlike caste or ethnicity, migrant identity in China is administratively created to control spatial
labor mobility and reinforced through merely decades of differential administrative treatment of
rural-urban residents. We, therefore, have an opportunity to test whether systemic disparity
between households over a short period of time (compared to thousands of years for caste and
ethnic disparities) can affect performance under incentives. This provides us with more powerful
evidence on the impact of institutions and social exclusion in creating identities. Finally, as a policy
3
intervention in the domestic labor market, China’s hukou system was initially designed to prevent
potential problems that may be caused by massive rural-urban labor migration. By focusing on
migrant identity created by such a policy this paper carries broader implications for policy making
in other economies where regional labor migration or global immigration may pose challenges in
the labor markets.
Our results indicate that when hukou identity is made salient, the performance of rural
migrant students declines substantially by 11.96 percent in the overall sample, and by 17.57 percent
under piece rate payments, compared to local urban students. In addition, migrant students’ average
ranking of experimental earnings declines by 13.5 percentiles relative to local urban students under
piece rate payments. These effects are, however, insignificant under the tournament regime,
suggesting that competition may mitigate the debilitating effect of the salience of migrants’ inferior
identity.
Our findings indicate that identities created by the hukou system may affect individuals’
economic performance but its impact varies by the payment regime. The negative impact of hukou
salience under piece rate payments suggests that hukou identity salience may exacerbate the
widening income gap between migrants and non-migrants in urban China (Ravallion and Chen,
2007). Policy interventions based on categorization of citizens and differential treatments by those
categories may lead to unintended effects on individuals’ economic behavior, and, thereby, have
wider economic consequences.
There exists a large literature on the impact of the hukou system on children’s health and
educational outcomes in rural China. Our research extends this literature to the relatively
unexplored urban context with two broad implications.1 First, existing literature suggests that high
Studies suggest that educational attainment of potential migrants in rural areas is lowered due to occupational
segregation in cities (deBrauw and Giles, 2008); the health and educational attainment of children is adversely affected
1
4
levels of inequality may have adverse implications for labor productivity (Cohn et al., 2011). This
in turn may have adverse consequences for sustaining economic growth (Berg, Ostry, and
Zettelmeyer, 2012; Berg and Ostry, 2011). Our results, thus, highlight the importance of avoiding
discriminatory social and economic policies in order to ensure inclusive and sustainable economic
growth in developing countries. Second, the results underline the need for more research on the
impact of social exclusion on individual behavior and the effectiveness of existing redistributive
policies that aim at reducing social exclusion. For instance, while affirmative action has generated
tremendous policy debate, there is limited evidence of its impact on social cohesion as well as
economic efficiency.
The remaining sections of the paper are organized as follows. Section 2 places this study in
the context of existing literature. Section 3 provides a brief background on China’s hukou system.
Section 4 describes the experimental design and the data. The results of the analysis and its
discussion are presented in Section 5. Section 6 concludes.
2.
Literature Review
The importance of incorporating social identity into economic analyses is stressed by Akerlof and
Kranton (2000). An expanding theoretical (Shayo, 2005; Fang and Loury, 2005; Benabou and
Tirole, 2007; Horst, Kirman and Teschl, 2006) and a growing number of experimental economics
studies show that social identity has an impact on preferences and economic behavior (Eckel and
Grossman 2005; Goette, Huffman and Meier 2006; Charness, Rigotti and Rustichini 2007; Chen
and Li 2009; Benjamin, Choi and Strickland, 2010; Benjamin, Choi and Fisher, 2013; Chen and
Chen 2011). To the best of our knowledge, only two studies investigate the impact of social identity
due to the absence of parents who have migrated to the cities (Lee and Park, 2011; Meng and Yamauchi, 2012). Unlike
this existing literature the focus of our study is migrant children in urban areas.
5
on economic outcomes in developing countries experimentally. Hoff and Pandey (2006, 2012) find
that social identity – a product of history and culture – shapes one’s belief system and has a
pronounced impact on an individual’s performance under incentives. They show that making caste
salient to middle school male students in rural India lowers the performance of low-castes relative
to high-castes even when rewards for performance depend solely on individual effort. Hoff and
Stiglitz (2010) discuss why ideologies of social unworthiness, as cognitive frames, can be so
powerful.
Concerns about rising inequality accompanying rapid economic growth have been growing
in recent years. In China, income inequality has risen significantly since economic reforms were
initiated in 1978 (Ravallion and Chen, 2007; Yao, Zhang and Hanmer, 2004; Chen, 2002; Kanbur
and Zhang, 1999). Seventy percent of the overall income inequality in China was accounted for
by rural-urban inequality during 1983-1995. Further, the contribution of intra-urban inequality to
overall inequality increased by 96 percent over the same period (Kanbur and Zhang, 1999). Behind
these inequality statistics lie rapid urbanization (Ravallion and Chen, 2007) and the accompanying
labor migration from rural to urban areas – the largest in world history. Because of the hukou
institution, migrant workers, most being unskilled rural laborers, are treated as outsiders with
limited access to economic resources and opportunities (Liu, 2005).
Research suggests that on average those with a rural hukou are socio-economically worseoff than those with an urban hukou in China. 2 However, due to confounding unobservable
individual characteristics, causality between self-perceptions of hukou status and economic
Liu (2005) finds that compared to urban residents those who obtained urban hukou later in their lives have
significantly lower educational attainment and healthcare benefits. Lu and Song (2006) find that local urban workers
earn substantially higher hourly wage than those without local urban hukou. Whalley and Zhang (2007) show that
removing hukou barriers to domestic labor mobility will significantly reduce inequality and increase economic
efficiency gains.
2
6
behavior is hard to establish using survey data or direct field observations. Hence, the current
literature almost entirely focuses on restricted labor mobility and discrimination in resource
allocation in the hukou system to explain rural-urban and intra-urban economic inequality (Liu,
2005; Lu and Song, 2006; Whalley and Zhang, 2007).3
3.
The Hukou System and Social Identity in China
The modern-day household registration or hukou system evolved gradually following the success
of the communist revolution in China in 1949. Under this system every citizen was legally bound
to register her or his single permanent place of residence (hukou suozaidi) say city x in province y,
and the type of hukou (hukou leibie) which was either agricultural (rural) or non-agricultural
(urban). Strict controls were imposed on mobility of rural hukou holders to urban areas,
perpetuating discrimination against them in several ways. Employment opportunities favored
urban hukou holders in cities (Chan and Zhang, 1999). Urban hukou holders, but not rural hukou
holders, were eligible for the ration stamps that guaranteed subsidized products (Liu, 2005). Urban
residents (particularly employees of the state-owned enterprises) were granted generous fringe
benefits including subsidized housing, health services and education. In contrast, such benefits to
rural residents were provided by their communes or villages, which were usually of inferior quality
and of highly varying reliability (Lin, Cai and Li 1996; Cheng and Selden, 1994). A household’s
hukou was (and continues to be) inherited by the next generation. Hukou status and thereby
educational attainment and eventually employment opportunities of the next generation of rural
populace was restricted by birth.4
Some recent studies use survey data on both migrants and local residents in urban China to explain earnings
differentials based on the observed characteristics of the two groups (Qu and Zhao, 2012).
4
Before 1998, children of rural-urban marriages had to follow the mothers’ hukou type (Chan and Zhang, 1999). They
are now allowed to inherit either the fathers’ or the mothers’ hukou.
3
7
Following China’s transition from a centralized to a market economy starting in the late
1970s, the number of people migrating in search of jobs surged after market reforms (and abolition
of communes) and an easing of government regulations on spatial migration. The past three
decades have, thus, witnessed a drastic increase in the number of temporary non-local hukou
residents (either registered or non-registered) in urban centers. However, these migrants are not
entitled to urban benefits unless their hukou is converted to a full urban one. Rural-urban hukou
conversion is possible but only through very limited channels.5 Thus the hukou system transitioned
from an institution of direct to indirect control over spatial migration.
As of when this study was conducted, the following major and persistent gaps between rural
migrants and urban residents existed in large urban centers: (1) labor market and occupational
segregation: employment in government offices and state-owned enterprises in cities continued to
be unavailable to rural migrant workers unless they converted to an urban hukou; (2) lack of social
insurance and social welfare benefits for rural migrant workers such as unemployment and health
benefits. In addition, government subsidized low-rent housing in large cities is only available for
local urban hukou holders.6
It is important to realize that these factors, along with the resulting social segregation of
migrants, suggest that the experience of a rural migrant in China is not transitory as in other
developing countries (Chan, 1996; Solinger, 1999). Geographical control of rural populace until
Possible channels for rural-urban hukou conversion include recruitment by state-owned enterprises, enrollment in
higher education institutions, and land acquisition by government (Chan and Zhang, 1999). Since 1992 urban hukous
are also sold at a high fee to eligible investors, property buyers and professionals.
6
On July 30, 2014, China’s government issued a circular announcing that it would reform the hukou system
(http://www.gov.cn/zhengce/content/2014-07/30/content_8944.htm). New government regulations will remove the
limits on hukou registration in townships and small cities, relax restrictions in medium-size cities, and set
qualifications for registration in big cities. However, for mega-cities with more than five million residents (including
Beijing), the government will continue to enforce strict control on the influx of migrants
(http://www.chinadaily.com.cn/china/2014-07/30/content_18216278.htm).
5
8
the 1970s has been replaced by economic and social divisions between rural and urban hukou
holders within the urban areas of China.
The Hukou System and Schooling in Urban China
Chinese citizens are entitled to subsidized public education only in the area of their legal permanent
residency. In most cities non-local hukou holders cannot enroll their children in local schools
unless the schools have quotas for ‘guest’ students. These ‘guests’ usually have to pay higher fees
than local hukou holders, and the fees can be a significant proportion of migrants’ incomes (Xinhua
News Agency, July 10, 2002). 7 Slum schools built by migrant workers exclusively for their
children are typically opposed by local authorities, or stringent physical and financial requirements
are imposed on them to obtain a legal status.
During China’s urbanization process, the newly developed urban areas contain both nonagricultural and agricultural population because of the hukou classification system (Chan and
Zhang, 1999). Thus the population in large municipalities such as Beijing and Shanghai usually
consist of four different hukou categories: local urban (residents of the urban areas of municipality),
non-local urban (migrants who are urban residents of less-developed cities), local rural (residents
of the rural areas of municipality), and non-local rural hukou holders (migrants from rural areas of
provinces outside municipality). The local urban residents are considered to be at the top of the
social hierarchy while the migrants from rural areas are typically at the bottom. But the comparison
of socio-economic status between local rural and non-local urban is not clear. Due to this ambiguity
According to the report, “..... These (migrant) schools teach between 20 and 3,000 students each and charge about
¥300 (36.1 US dollars) each term. When migrant children go to public schools, their parents have to pay ¥500 (60.2
US dollars) in tuition fees each term, plus ¥1,000 (120.5 US dollars) for selection of the school and ¥1,000 to ¥30,000
(3614.5 US dollars) as sponsorship.”(http://www.china.org.cn/english/China/36594.htm).
Migrant laborers in Chinese cities earn an average of ¥966 per month in 2006 according to a National Bureau of
Statistics Survey Report. (http://english.peopledaily.com.cn/200610/22/eng20061022_314208.html).
7
9
and the limited number of students in these two categories in our sampled schools, we excluded
the local rural and non-local urban hukou holders from this study, and focused on the two sharply
disparate groups. Since our experiment was conducted in Beijing, this study includes Beijing urban
(hereafter High) and non-Beijing, rural (hereafter Low) hukou holders.
4.
Experimental Design and Implementation
Our experiment adopts the design of Hoff and Pandey (2006, 2012). We manipulate hukou salience
– subjects’ hukou identity is made salient and public in the identity salience treatment, and is kept
private in the control treatment. We also vary the payment regime by using piece rate and
tournament in each of the treatments.
Incentivized cognitive task
The experiment was conducted using paper and pencil in a
standard classroom setting with six subjects (3 H and 3 L types) seated at separate desks with fairly
large distance from one another. Randomly assigned subject ID numbers were used to ensure
anonymity of decisions throughout the experiment. Before the experiment started participants were
greeted by a female experimenter and each paid 3 Chinese yuan (¥3) participation fee upon arrival.
The experimenter then explained the tasks and rules. We used level-2 maze puzzles from Yahoo!
games (Gneezy, Niederle, and Rustichini, 2003; Niederle and Vesterlund, 2007; Hoff and Pandey,
2006). The task was to find a path through the field from one side to the other without crossing the
solid lines (Appendix A). The experimenter explained the rules using a simple maze, and showed
how to solve another one of similar difficulty level as those used in the experiment. 8 Subjects were
given five minutes to practice with an additional maze, then participated in two 15-minute rounds
of experiment. In each round, they were given a booklet of 15 mazes, and had up to 15 minutes to
The instructions are adopted from Hoff and Pandey (2006, 2012). The English translation of the experimental
instructions is included in Appendix B.
8
10
solve as many as possible. All thirty mazes were of identical difficulty level.
The homogeneous reward system (hereafter the Pure Piece Rate (PP) treatment) used piece
rate compensation in both rounds – subjects were rewarded with ¥1 for each maze solved correctly.
The heterogeneous reward system (hereafter the Mixed Tournament (PT) treatment) consisted of
piece rate in the first round (¥1 per maze), and tournament in the second round in which only the
winner (who solved the most number of mazes in the session) was rewarded with ¥6 per maze and
other subjects received zero.9 In the case of a tie, each of those who solved the highest number of
mazes was rewarded with ¥6 per maze.
Subjects were told that the task consisted of two rounds. But the instruction for the second
round, including the payment scheme, was not given until after round one. Therefore, the payment
structure was identical in round one for the PP treatment and the PT treatment. At the end of each
round, maze booklets were collected and left outside the classroom by the experimenter for the
graders. Since the experimenters did not leave the classroom at any point of time during the session,
subjects were aware that their mazes were not graded by their experimenters. A survey was
conducted at the end to collect demographic information. Thereafter, the grading results were left
outside the classroom and picked up by the experimenters. Subjects were then informed about their
performance (and results of the tournament, if applicable), paid individually in private, and
dismissed. Throughout the process, neither the subjects could see the graders nor the graders could
associate the booklets with individual subjects.
Identity manipulation
There were two treatments in our experiment – the identity salience
treatment and the control treatment. The identity salience treatment differs from the control
treatment by manipulating the salience of one’s hukou identity before the incentivized tasks.
9
To avoid zero payoffs we did not use tournament in both rounds.
11
Priming, a technique often used in psychology, introduces certain stimuli (‘primes’, including
image, audio, or text such as a questionnaire and an article) to activate subjects’ knowledge of
social structures. As shown in a large literature in psychology (see Bargh, 2006 for a review) and
a few recent economic studies (Hoff and Pandey, 2006, 2012; Benjamin, Choi and Strickland, 2010;
Benjamin, Choi and Fisher, 2013), priming social identities may influence behavior and attitudes.
In this study, we made hukou identity salient in the identity treatment by using a survey and
publicly verifying subjects’ hukou identity at the beginning of the experiment, while in the control
treatment subjects’ hukou identity was kept private and not primed. In the pre-experiment survey,
used to prime hukou identity (see Appendix C), subjects were asked where they were born, whether
they spoke Beijing dialect at home, whether they (their classmates or teachers) considered
themselves (them) as a Beijing local, and how much miscellaneous fees they were charged by the
school at the beginning of the semester. They were also asked to compare Beijing local students
with ‘guest’ students on diligence in academic study, class participation, extracurricular activities,
and attitude towards money in daily spending. After the survey, individuals’ hukou was publicly
verified by the experimenter along with other factual information in the following order: name,
date and month of birth, and hukou.10 This procedure simulated some real life scenarios that these
students experience in school. For example, to determine the amount of miscellaneous fees due,
students’ hukou status needs to be verified, sometimes in public.11 In the control treatment, the
sessions started with neither the survey nor public verification of hukou, and proceeded directly to
the incentivized cognitive task.12
The experimenter said in public, “According to the information from the school’s registrar’s office, your hukou
belongs to [province/city] and it is a(n) [agricultural/non-agricultural] hukou.” Student’s year of birth was not
revealed to avoid affecting their self-confidence given that they were from different grades.
11
Unlike in many western countries, date of birth, students’ grades, and similarly hukou type are not treated as
confidential information in Chinese culture.
12
Note that there is a subtle difference in the experimental design, and therefore the interpretation of findings, between
our study and Hoff and Pandey (2006, 2012). In the latter study all subjects’ castes were verified privately before they
10
12
The experimental intervention (survey and public verification) might have had effects on
subjects that are unrelated to hukou salience. For instance, subjects’ attention may have been
diverted from (viz. by responding to a survey, independent of its content) or, conversely, drawn to
(viz. by calling out their names) the maze task. Such direct effects could be incorrectly construed
as the impact of hukou salience on performance if we focus solely on the difference between the
control and salience treatments for each hukou type. Fortunately, our experimental design allows
any such effect to be differenced out empirically when we compare the effect of the salience
treatment on the L subjects to its effect on the H subjects. We, therefore, follow a difference-indifference empirical strategy to measure the causal impact of hukou salience by comparing the
effect of the experimental intervention on the L subjects’ performance to that on the H subjects’.
Selection of subjects
Subjects in our experiment were 8-12 years old students, randomly
selected from four elementary public schools in Beijing. These schools suited the requirements of
our experiment on several fronts. On average, one-third of enrolled students in the sampled schools
had a non-Beijing hukou. The proportion of students from migrant families was comparable to the
migrant population of the Beijing metropolitan area, and these schools were located in districts
where per capita GDP was comparable to the average in the Beijing municipal area. We obtained
individual hukou records, in advance, from the schools’ registrar’s offices for assigning subjects to
the experimental sessions.
The study focuses on primary school students rather than adults for several reasons. First,
focusing on young individuals allows us to avoid using selective samples of people whose decision
were assigned to different treatments. Hence caste identity was made privately salient irrespective of the treatment
condition. In our study subjects’ hukou identity was made privately salient through the pre-experiment survey only in
the salience treatment. This was followed by public verification in the salience treatment as discussed above. Thus
any identity effects we observe would be attributable to the overall salience of hukou identity - public as well as private.
13
to migrate to cities may be systematically correlated with their abilities and intrinsic preferences.13
Second, using adult subjects may introduce potential confounds since their accent and clothing
usually provide clues to where they originate from and give away their hukou identity even in the
absence of the hukou identity prime. Since our subjects wore identical school uniforms and,
relative to adults, the young tend to pick up local accents easily, the choice of subject pool ensured
that it was difficult to observationally infer one’s hukou type. Third, a study that identifies potential
adverse impact of the hukou system on young migrants would suggest the importance of policy
interventions starting at young ages. Although our sample was young, almost all participants
understood what their hukou type was. Last but not least, school children of this age are expected
to respond to monetary incentives given that they tend to receive pocket money or allowances from
their parents.14
Subjects were randomly selected from among 3-6 graders at three schools and among 3-5
graders in one (see details on the random sampling and assignment procedure in Appendix E.)
Each session of the experiment consisted of six subjects of the same gender with three from each
of the H and L hukou types. To minimize the probability that students knew one another a priori,
we obtained student rosters and stratified students by gender, grade, class section, and hukou type.
Three students of the same hukou type were randomly selected, each from a different grade. Two
students from the same grade (with different hukou types) were randomly selected, each from a
different class section. The sessions in both treatments were formed in the same way. Consent for
student participation was obtained from their teachers. All selected students were informed, before
One may be concerned that since parents of children who migrate to urban areas are a selective sample, the L type
could still be subject to this bias. Even if this were the case, the potential selection should not affect the internal validity
of our experimentally identified effects since our treatments are randomized.
14
A study conducted jointly by the China Youth and Children Research Center (CYCRC) and the Bank of China (2010)
shows that 47.1 percent of primary school age children living in urban areas receive pocket money for doing household
chores (http://www.cycs.org/Article.asp?Category=1&Column=444&ID=16893). Liu (2011, chapter 4) also discusses
sources of children’s pocket money.
13
14
the experiment started, that they were free to choose to or not to participate in the experiment.
However, all selected students chose to participate. In some instances, sessions with fewer than 6
students occurred when a selected student happened to be absent from school on the day of the
experiment.
The experiment was conducted in May and December 2007, and December 2008. At each
school the experiment included four treatments including the PP and PT control treatments as well
as the PP and PT identity salience treatments. Table 1 summarizes the features of experimental
sessions. Experimental sessions were conducted separately for boys and girls during class breaks
on school days. In total, we conducted 72 sessions (418 subjects) with mixed hukou types,
including 61 full sessions (40 male and 21 female sessions with 3H and 3L subjects each) and 11
sessions with fewer than 6 subjects. Average earnings per subject were ¥19.70 in the control
treatment and ¥19.60 in the identity salience treatment, about three times the average weekly
allowance that these students received from their parents.
As mentioned above, pre and post-experiment surveys were conducted to elicit subjects’ selfperceptions and demographic characteristics. 15 The summary statistics of the pre-experiment
survey to prime identity in the salience treatment (Table 2) show that H subjects were both more
likely to have been born in Beijing and lived longer in Beijing if born elsewhere, relative to L
subjects (rows 1-2). The L subjects were less likely to speak in the Beijing local dialect at home
(row 3). Their self-image was highly correlated with their hukou type: only 9.3 percent of L
subjects (compared to 86.4 percent of H subjects in row 4) considered themselves Beijing locals;
only 23.1 percent of L subjects compared to 81.7 percent of H subjects expected to be considered
Beijing locals by their fellow students (row 5), while 23.6 percent of L subjects compared to 90
The survey questionnaires are included in Appendices C and D. Appendix F summarizes the responses, by students’
hukou type, from the post-experiment survey in both treatments.
15
15
percent of H subjects had a similar expectation of their teachers (row 6). As shown in the last
column of Table 2, all these comparisons between L and H subjects are statistically significant at
1 percent level. Subjects were also asked about their perceptions of local and migrant students on
four aspects, including how hard they worked, their involvement in classroom and extra-curricular
activities, and attitude towards money (rows 8-15). Take rows 8-9 as an example, 16.4 percent of
L subjects and 13.5 percent of H subjects thought that Beijing local students were more
hardworking than their rural migrant counterparts while 32.7 percent of L subjects and 36.9 percent
of H subjects thought the opposite. It shows both L and H subjects shared very similar perceptions
(p = 0.552 in row 8, p = 0.511 in row 9.) In addition, the proportion of L subjects (H subjects alike)
who thought migrant students were more hardworking significantly exceeds those who thought
Beijing local students were more hardworking (32.7 percent in row 9 vs. 16.4 percent in row 8 for
L, p = 0.005; 36.9 percent in row 9 vs. 13.5 percent in row 8 for H, p = 0.000). These patterns
manifest themselves in subjects’ responses to other perception questions (rows 10-15, with only
one exception in row 10), which shows that migrant students were generally perceived to be more
hardworking, less active participants in class and in extracurricular activities, but more careful
with money, compared to their Beijing local counterparts.16 These perceptions are not surprising
given the relatively disadvantaged background of migrant families as indicated by the summary
statistics from the post experiment survey in Appendix F: parents of the migrant families were less
likely to be employed in private or public enterprises (considered prestigious in Chinese society,
as opposed to being non-salaried and self-employed) and were less likely to have college or
graduate degrees, all statistically significant differences at 1 percent level as shown in rows 1-4 in
Appendix F. The perceived own academic performance is slightly, but insignificantly, poorer by
In row 10, 24.5 percent of L subjects, compared to 41.4 percent of H subjects, thought that Beijing local students
were more active participants in class (p = 0.008.)
16
16
the L subjects than by the H subjects (row 5 in Appendix F, p = 0.149). These survey responses
suggest that the L subjects may have perceived themselves as inferior to the H subjects, confirming
the negative stereotype of rural migrants in Chinese society.
Table 3 presents subjects’ characteristics by hukou type and treatment. For each hukou type
it shows insignificant differences across treatment in almost all observable characteristics,
including participants’ average age, grade, proportion born in Beijing, number of years of local
residence, prior experience with the maze game, prior acquaintance with other participants, the
amount of pocket cash given by parents, and their parental characteristics such as education and
occupation.17 This indicates successful randomization of subjects across treatment. In addition, we
run an F test of the hypothesis that these observable characteristics, listed in Table 3, are jointly
zero by regressing the salience treatment dummy on these variables, including a dummy for
missing values. The P-values of the F test are 0.316 for H and 0.298 for L subjects, so the null
hypothesis that these variables are jointly zero cannot be rejected for either hukou type. In the
regression analysis in section 5, we also conduct a robustness check of our results by including
these individual and parental observable characteristics as additional control variables.
5.
Results
In this section, we investigate whether and how the activation of hukou identity salience influences
individual’s performance in the maze games. We first present the descriptive statistics and then
discuss the regression analysis, which incorporates individual and school characteristics. The
analysis excludes 11 sessions which had fewer than 6 subjects due to school absence of pre-
Marginal significance is found in 3 out of 22 across-treatment comparisons in Table 3. Specifically, the proportion
of H subjects who were born in Beijing was marginally higher in the salience treatment than in the control treatment
(p = 0.056). The proportion of L subjects who had prior experience with the maze games before the experiment was
marginally higher in the control treatment than in the salience treatment (p = 0.054). H subjects in the control treatment
received slightly more pocket money than those in the salience treatment (p = 0.069).
17
17
selected subjects on the day of the experiment. We, therefore, have 366 subjects in our sample.
Robustness checks, including analysis based on the entire sample, are conducted and discussed
later in this section.
Recall that the Pure Piece Rate (PP) treatment consists of piece rate in both rounds whereas
the Mixed Tournament (PT) consists of piece rate in round 1 and tournament in round 2. In the
analyses, we first show results by pooling the data across payment methods (piece rate and
tournament) and gender groups. We then split the sample on one dimension, either by payment
(with piece rate including both rounds of PP and the first round of PT, and with tournament
including only the second round of PT) or gender. We further split the sample on two dimensions,
by payment and gender group.
As discussed in section 4, since both L and H subjects were subject to the pre-experiment
intervention, our experimental design allows us to empirically difference out any potentially
confounding effects of the experimental intervention (which are unrelated to hukou salience) by
comparing the treatment effects across the two hukou types. Therefore, our discussion of results
will focus on the differential effect of the experimental treatment on L and H subjects, i.e., the
difference-in-difference estimates. For the sake of completeness and ease of interpreting the
difference-in-difference estimates, we will also discuss single difference estimates, i.e., the
treatment effects on L and H, separately. However our conclusions on the causal impact of the
salience of hukou identity will be drawn based on the difference-in-difference estimates.
Figure 1 shows the average number of mazes solved by treatment and hukou type. Panels 1a1c in the first row describe the data for all payment methods pooled, piece rate, and tournament,
respectively. Panels 2a-2c in the second row and 3a-3c in the third row describe the data in the
18
same sequence for boys and girls separately. 18 As shown in Panels 1a-1c (with the summary
statistics detailed in Table A1 of Appendix G), when hukou is made salient, L’s performance
worsens by 0.528 mazes in the pooled sample (p = 0.071) and by 0.708 in piece rate (p = 0.028),
but slightly improves by 0.181 in tournament (p = 0.756), compared to the control treatment. H’s
performance in the salience treatment, on the other hand, improves by 0.369 mazes in the pooled
sample (p = 0.250) and by 0.594 in piece rate (p = 0.089) but worsens by 0.173 in tournament (p
= 0.792), compared to the control treatment. Therefore, we observe a significant decrease in L’s
performance, relative to H’s, due to hukou salience in the pooled sample (0.897, p = 0.039) and in
piece rate (1.302, p = 0.006), and an insignificant increase in L’s relative performance in
tournament (0.355, p = 0.686). Disaggregating the sample by gender, Panels 2a-2c, 3a-3c and the
difference-in-difference results in the lower panel of Table A1 show that when hukou is made
salient as opposed to the control treatment, both L boys and L girls perform worse relative to their
H counterparts in the pooled sample (-0.661 for L boys, p = 0.238; -1.102 for L girls, p = 0.087)
and in piece rate (-1.051 for L boys, p = 0.091; -1.641 for L girls, p = 0.017), but both perform
insignificantly better than their H counterparts in tournament (0.504 for L boys, p = 0.645; 0.667
for L girls, p = 0.621).
Figure 2 focuses on the comparison of experimental earnings distributions by treatment. In
each payment regime and treatment, subjects, regardless of their gender and hukou background,
are ranked based on their experimental earnings (excluding the ¥3 participation fee). The
cumulative distribution functions (CDFs) of the ranking percentiles are then plotted by treatment
in the pooled sample (Panels 1a-1b), in piece rate (Panels 2a-2b), and in tournament (Panels 3a3b) for each hukou type. In Panels 1a-1b of the pooled sample, we find that compared to the control
Summary statistics are reported in the upper panel of Table A1 in Appendix G. The treatment effects on L and H as
well as the difference-in-difference estimates, along with p values, are reported in the lower panel of Table A1.
18
19
treatment, the H subjects’ average ranking in earnings distribution increases by 4.99 percentiles (p
= 0.169), and the L’s average ranking decreases by 4.29 percentiles (p = 0.195) in the salience
treatment, leading to a marginally significant 9.28-percentile decrease in L’s relative ranking (p =
0.059). In Panels 2a-2b (piece rate), the average ranking in earnings distribution increases by 6.82
percentiles (p = 0.065) for H but decreases by 7.42 percentiles (p = 0.031) for L in the salience
treatment, leading to a significant fall in L’s relative ranking by 14.24 percentiles (p = 0.005),
compared to the control treatment. Interestingly, in Panels 3a-3b (tournament), the average ranking
decreases by 2.84 percentiles for H (p = 0.734) but increases by 2.85 percentiles for L (p = 0.633),
therefore L’s relative ranking increases insignificantly by 5.69 percentiles (p = 0.579) in the
salience treatment compared to the control treatment. These observations confirm those in Figure
1.
5.1 Effects of Hukou Salience
We next use the following school fixed-effects model to analyze the impact of hukou salience on
individual performance:
𝑌𝑖𝑠 = 𝛽0 + 𝛽1 𝐿𝑖𝑠 + 𝛽2 𝑠𝑎𝑙𝑖𝑒𝑛𝑐𝑒 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖𝑠 + 𝛽3 (𝐿𝑖𝑠 ∙ 𝑠𝑎𝑙𝑖𝑒𝑛𝑐𝑒 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖𝑠 ) + 𝛤𝑋⃗ + 𝛼𝑠 + 𝜂𝑖𝑠
The dependent variable of interest (Yis) is the number of mazes solved by individual i in school s.
The independent variables include low hukou type (L) (H type is in the omitted category), dummy
variables for the identity salience treatment (salience treatment), and their interaction to allow the
treatment effects to differ by hukou type. The coefficient estimate 𝛽1 measures the difference in
performance between L and H in the control treatment. 𝛽2 and (𝛽2 + 𝛽3 ) measure the treatment
effects on the performance of H and L subjects, respectively. As discussed in section 4, 𝛽3, the
difference-in-difference of the treatment effects between L and H, identifies the causal effect of
20
the salience of hukou identity, and hence is the main coefficient of research interest. The control
variables in ⃗X⃗ include gender (male), grade, whether the student had played similar or the same
maze games before. In the analysis we also include round or payment regime specific effects (i.e.,
round 2 of piece rate or tournament) wherever appropriate. 19 School fixed effects, αs, capture
unobservable, school-specific characteristics. ηis is the error term. Standard errors are clustered at
the individual level.20 In the bottom panel of the result tables, we also report the treatment effect
on L, i.e.,(𝛽2 + 𝛽3 ), and tests of equality of coefficients across specifications.
Table 4 focuses on individual performance in the maze game. Column 1 shows the results
for the pooled sample. We find that while the salience treatment has an insignificant effect on the
H type’s performance (𝛽2 = 0.167, p = 0.643), it significantly reduces L’s performance (𝛽2 + 𝛽3 =
-0.714, p = 0.018, row (a) of the bottom panel). As a result, the L type perform worse by 0.880
mazes (𝛽3 , p = 0.059) relative to the H type when hukou identities are made salient. Given that the
L type solved 7.361 mazes on average in the control treatment for the pooled sample, the result
suggests that identity salience leads to an 11.96 percent decrease in the L type’s performance
relative to H’s.
In piece rate (column 2), L’s performance significantly decreases by 0.913 mazes (p = 0.005,
row (a) of the bottom panel), and H’s performance insignificantly increases by 0.329 (p = 0.381)
in the salience treatment compared to the control treatment. Thus the L type’s performance worsens,
relative to H’s, by 1.241 mazes (p = 0.012) when hukou identities are made salient. Compared to
L’s average performance of 7.065 mazes in the control treatment in the piece rate regime, this
There is a significant correlation of 0.7 between grade and subjects’ age. Using age or grade does not affect the
results. In the pooled analysis, fixed effects of round 2 for piece rate and tournament are included separately. Additional
analysis in Table A2 of Appendix G shows no significant differences in the experimental effects (𝛽2 and 𝛽3 ) between
the two rounds in PP, or between round 1 of PP and round 1 of PT.
20
Using the same specification, we also analyze effects on two additional outcomes, the likelihood of solving the most
mazes and experimental earnings. These results are reported in Tables A6 and A7 of Appendix G.
19
21
impact is equivalent to an economically substantial decrease of 17.57 percent in L’s relative
performance in piece rate due to hukou salience. In tournament (column 3), the treatment effects
on the performance of H and L types (-0.258, p = 0.667 for H; -0.114, p = 0.823 for L) and the
difference-in-difference estimate 𝛽3 (0.144, p = 0.857) are all statistically insignificant. In addition,
row (d) shows that the difference-in-difference estimate 𝛽3 in tournament (0.144 in column 3) is
marginally significantly different (p = 0.084) from its counterpart in piece rate (-1.241 in column
2). This interesting result, further discussed at the end of this section, suggests that the negative
impact of hukou salience on L observed in the piece rate regime may be mitigated by competition
among subjects in tournament.
The results by gender in columns 4-5 in Table 4 show no statistically significant differences
in the treatment effects or the difference-in-difference estimate 𝛽3 across gender in the pooled
sample (rows (b)-(d) of the bottom panel). Columns 6-9 display results disaggregated by gender
and payment regime. In the piece rate regime (columns 6-7), the identity salience treatment leads
to an insignificant increase in performance by 0.139 for H boys (p = 0.780) and 0.643 for H girls
(p = 0.256), while it reduces the performance by 0.823 for L boys (p = 0.055) and 1.043 for L girls
(p = 0.030). Hence, the impact of identity salience on subjects’ performance, measured by 𝛽3 , is
-0.963 (p = 0.139) for L boys and -1.686 (p = 0.024) for L girls, relative to their H counterparts. In
tournament (columns 8-9), the treatment effects and the difference-in-difference estimate 𝛽3 are
statistically insignificant for both boys and girls.
Comparing results across columns 6-9, a few observations emerge. First, conditional on the
payment regime, we again find insignificant differences in both the treatment effects and the
difference-in-difference estimate 𝛽3 across gender groups (columns 6 vs. 7, 8 vs. 9, rows (b)-(d)).
Second, conditional on gender, we find that the treatment effects and the estimate for 𝛽3 are not
22
statistically different across payment regimes (columns 6 vs. 8, 7 vs. 9, rows (b)-(d)) in all cases
with one exception – the treatment effect on L girls’ performance in piece rate is significantly
different from that in tournament (-1.043 in row (a) of column 7 vs. 0.780 in row (a) of column 9,
p = 0.025 in row (c)). We also observe that the impact of hukou identity salience on L’s relative
performance ( 𝛽3 ) changes from negative in piece rate to positive in tournament (from -0.963 to
0.374, p = 0.184 for boys; from -1.686 to 0.278, p = 0.143 for girls). Although these changes are
statistically insignificant, they echo our previous observation on the marginally significant
difference in 𝛽3 between the piece rate and tournament regimes (-1.241 in column 2 vs. 0.144 in
column 3, p = 0.084 in row (d)). These observations again could imply that the negative impact of
hukou identity salience on the L type in piece rate might be mitigated by competition in the
tournament regime.21
Overall, results in Table 4 indicate that making hukou identity salient reduces L’s
performance, relative to H’s, marginally significantly in the pooled sample and significantly in the
piece rate regime, while this negative impact is absent in the tournament regime. This implies that
one’s hukou identity, when made salient, may influence individual’s performance on incentivized
tasks depending on the payment scheme.
What does this impact entail regarding the income distribution for different hukou groups?
Across all columns in Table 4, the effects of other covariates are generally as expected. The large positive
coefficients of the ‘piece rate round 2’ and ‘tournament’ dummies suggests substantial improvement in performance
over time. Boys solved more mazes than girls as shown in columns 1-3. The higher the grade, the better is the
individual’s performance. Previous experience with similar or the same maze games does not increase one’s
performance in most cases. Additional analyses, as robustness checks, are conducted and presented in Appendix G:
Table A3 reports analysis based on the entire sample (rather than the full sessions). Table A4 reports analysis by adding
other control variables, such as the number of others known in the session, pocket money received, parental education
and occupation. In Table A5, we disaggregate the data by session type (i.e., PP vs. PT) rather than payment regime
(i.e., piece rate vs. tournament). Table A6 reports logit analysis on the likelihood of solving the most mazes in the
session. We find that our main results hold throughout except that in Table A6 the impact of hukou salience becomes
insignificant in the pooled sample and marginally significant in the piece rate regime. The results in Table A6 are
statistically insignificant mainly because of the reduced statistical power when we use the likelihood to win as the
outcome measure. In addition, we include all the analyses based on the entire sample in Part II of the online appendix.
21
23
We investigate this question by examining the impact of hukou salience on the experimental
earnings gap between the H and L subjects in Table 5. Using the same empirical specification as
in Table 4, our dependent variable here is individuals’ earning percentile in an experimental round.
We find that in the pooled sample (column 1) the salience treatment insignificantly increases the
H type’s ranking (2.546, p = 0.550) and marginally significantly decreases the L type’s ranking (6.364, p = 0.064) in the distribution of earnings. Hence, making hukou identity salient results in a
decline in L’s ranking, relative to H’s, by 8.911 percentiles ( 𝛽3 , p = 0.105) in the earning
distribution in the pooled sample. In piece rate (column 2), hukou salience leads to a significant
13.500 percentile decline (p = 0.018) for L subjects, relative to H. In contrast, the estimate for 𝛽3 is
4.096 (p = 0.700) in tournament (column 3). Although 𝛽3 in tournament (4.096) is not significantly
different from that in piece rate (-13.50) (test of equality of coefficients in row (d)), the statistical
significance of this difference is just above the 10 percent level (p = 0.101). In addition, the impact
of identity salience by gender (columns 4 – 5) or by gender and payment regime (columns 6 – 9)
is largely in line with those in Table 4. Interestingly, the negative impact of hukou salience on L
girls’ relative ranking in piece rate (-19.52 in column 7) becomes positive and insignificant in
tournament (8.825 in column 9, p = 0.076 in row (d)). This result, albeit marginally significant,
might suggest that L girls’ performance may not be adversely affected by the salience of their
hukou identity under competition as opposed to under the piece rate regime. We revisit this point
in the discussions in subsection 5.2.22
Overall, results in Table 5 indicate that making hukou identity salient significantly decreases
the L type’s average ranking, relative to H’s, in the earnings distribution in the piece rate regime,
Results for another outcome variable ‘experimental earnings’ are reported in Table A7 in Appendix G. Note that the
results for piece rate (in columns 2, 6 and 7) are identical to those for piece rate in Table 4 since subjects received ¥1
per correct maze. There are no significant results for the pooled sample or for tournament, which might be due to the
lack of variations in earnings under tournament.
22
24
but the impact is again absent in tournament. These findings further confirm the previous results
on individuals’ performance.23
5.2 Discussion of Results
Our results indicate that individuals’ behavior is influenced by the hukou identity salience in terms
of their performance on incentivized tasks, under piece rate, and this impact may be absent when
competition is introduced. This suggests that the practice of permanently assigning households to
a rural status and discriminating against them for half a century has profoundly shaped citizens’
social identities, especially for the socially excluded hukou groups in China.
Our findings in the piece rate regime may be explained by stereotype threat, a wellestablished finding in social psychology and recently formalized in an economic model by Dee
(2014). The stereotype threat literature in psychology shows that making social identity salient in
the laboratory often makes subjects behave consistently with the stereotypes associated with that
social group, and hence may activate the negative stereotypes and hurt subjects’ performance in
relevant tasks. Most of these studies use tasks without economic incentives (Steele and Aronson,
1995; Aronson, Quinn and Spencer, 1998; Croizet and Claire, 1998). In China rural migrants are
generally stereotyped to be “uneducated, ignorant, dirty and having higher propensities to be
criminals” (Wang and Zuo, 1999). Migrant children are stereotyped to be less intelligent and have
low academic achievement. These negative stereotypes, confirmed by subjects’ response in the pre
In addition, we conduct a simulation exercise to analyze the possible impact of hukou salience on income inequality.
We first calculate the Gini coefficient using the subjects’ realized experimental earnings in the control treatment. We
then add the estimated treatment effects on H and L from our regression analysis (reported in Table A7) to their realized
earnings in the control treatment to generate the simulated earnings, and calculate the Gini coefficient of the simulated
earnings. The difference in the Gini coefficients can be inferred as a measure of the impact of hukou salience on
income inequality. The results show that when the hukou identity is made salient, the Gini coefficient increases by
0.0134 in the pooled sample (p = 0.810), increases by 0.0133 in the piece rate regime (p = 0.149), but decreases by
0.0847 in the tournament regime (p = 0.300). These findings are qualitatively in line with those reported in Figure 2
and Table 5. Details on the simulation exercise are included in Appendix H.
23
25
and post experiment surveys as discussed in Section 4, may be reinforced repeatedly and influence
economic performance by the migrant students. This self-image may be activated for migrant
students following the priming of their inferior hukou background in this study, and may prevent
them from performing to their full potential in the assigned tasks even in the presence of economic
incentives.
There is another possible explanation for the identity impact we observe: the intimidation
effect (Hoff and Pandey, 2006, 2012), i.e., knowing that they are evaluated along with their local
urban counterparts may hurt migrant students’ self-confidence in the identity salience treatment.
This is unlikely to be the driver of the results in this paper. Our subjects were randomly selected
from schools where migrant students and their local urban counterparts studied and interacted on
a daily basis. All the experimental sessions were conducted at subjects’ regular schools during
school days. Therefore, the intimidation factor, if any, is held constant across the control and the
identity salience treatments because subjects were randomly assigned to the sessions. Thus, the
findings based on the comparison across the two treatments should be net of the intimidation effect.
We further investigate any potential intimidation effect by comparing L’s performance in the
presence of H (in sessions with H and L subjects) with that in the absence of H (in sessions with
only L subjects). We conducted four, pure L hukou sessions with the PT identity salience treatment
(3 all-male and 1 all-female sessions). If a public announcement of hukou type in the presence of
H type truly intimidated the L type, then the performance of the L subjects in the mixed-hukou
sessions would have been worse relative to their performance in the pure-hukou sessions. However,
as shown in Table A8 of Appendix G, in both rounds of the PT salience treatment we find no
significant differences in the performance of the L subjects (girls, boys or pooled) between the
26
mixed and pure-hukou sessions. 24 We, therefore, conclude that our results are unlikely to be
explained by an intimidation effect.25
Recall the difference-in-difference estimates 𝛽3 in columns 2-3 of Tables 4 and 5 show that
making hukou identities salient results in a significant decrease in the L type’s relative performance
and their relative ranking in the earnings distribution in the piece rate regime, but this impact
becomes positive and insignificant in tournament. The difference in 𝛽3 between piece rate and
tournament is marginally significant (p = 0.084) in Table 4 and insignificant (p = 0.101) in Table
5. These findings, although not statistically precise, suggest that the impact of hukou salience may
differ between the two payment regimes, i.e., compared to the piece rate regime competition may
mitigate the adverse effects of negative stereotyping of migrants.26 Our conjecture is that the offer
of a strong incentive in tournament may have triggered substantially higher effort from the L type,
compensating for the negative shock on their performance due to the stereotype threat. This
conjecture is also discussed by Dee (2014), who suggests that individual performance may increase
For completeness, similar analysis of pure and mixed hukou sessions for the H type are reported in Table A9 in
Appendix G. Note that the statistical power of the comparisons in Tables A8 and A9 is limited because of the small
sample sizes.
25
Another possible explanation of our results is experimenter discretion, i.e., in our case, L subjects may expect to be
discriminated against by the experimenters, all of whom spoke Chinese Mandarin with an urban accent and hence
were perceived as Beijing locals. We discount this possibility. As pointed out in the discussion of experiment design,
only the experimenters knew the hukou types of the students in their sessions. The experimental procedure was
designed in such a way that neither the subjects could see the graders nor the graders could associate the booklets with
individual subjects. The experimenters did not leave the classroom throughout the session, so the subjects should have
been aware, particularly in round 2, that the experimenters were unlikely to influence the grading process. An
alternative design would have been to announce to the subjects in advance that the maze booklets would be graded by
helpers who would have no knowledge of who the subjects were. We did not choose this design in order to avoid
distracting subjects from their assigned task, and to minimize potential, invalid concerns with the grading process. To
illustrate, under piece rate if our results were driven by experimenter discretion we would observe a lower degree of
performance reduction (when identity is made salient) by L subjects in round 2 than in round 1, since in round 2 the
L subjects knew for certain that the experimenters were unlikely to influence the grading process. Nevertheless, we
observe the opposite, i.e., the reduction in performance by the L type is marginally greater in round 2 (1.044; p =
0.070) compared to round 1 (0.599; p = 0.066) when hukou identity is made salient. This suggests it is unlikely that
subject’s perception of experimenter discretion drives our results.
26
Note that the lack of the statistical precision in the comparison between piece rate and tournament may be due to
the smaller number of observations in tournament (174) relative to piece rate (558). Further research may help pin
down the impact of identity under different payment regimes.
24
27
in response to negative stereotype threat if an increase in effort substitutes for a negative ability
shock.
Another interesting question is whether the divergence in the impact of hukou salience, 𝛽3 ,
between piece rate and tournament exists for each gender group. We find that the impact of hukou
identity salience on girls’ relative ranking of earning distribution is marginally significantly
different between piece rate and tournament (p = 0.076, row (d), Table 5), but all other comparisons
of 𝛽3 between piece rate and tournament within each gender group do not show any significant
difference in Tables 4 and 5. Although a few studies (Oswald and Harvey (2000), Jamieson and
Harkins (2007), Fryer, Levitt, and List (2008)) also find an increase in performance by women
who are exposed to gender stereotype priming, research is limited on how social identity and
competition interact to affect economic behavior, and how the impact may manifest itself
differently for men and women. Unraveling the interplay among social identity, competition and
gender in economic decision-making is an interesting research agenda, but outside the scope of
this paper.27
6.
Conclusion
We conduct an experimental study to investigate the causal impact of social identity on individuals’
Table 4 also shows that the treatment effect on L girls’ performance significantly differs in piece rate from
tournament (-1.043 vs. 0.780 in row (a), p = 0.025 in row (c)) while this difference is absent for L boys, H boys and
H girls. Some readers may find it intriguing, since the gender gap literature shows that a competitive environment
enhances performance for men but not for women, compared to a non-competitive environment (Gneezy, Niederle,
and Rustichini, 2003; Gneezy and Rustichini, 2004). One possible explanation is L girls’ twin identities of being
female and having an inferior hukou identity. In Chinese society boys are preferred to girls and parents are likely to
invest more in boys’ education than girls’. This gender bias is more apparent in the rural areas than in the urban
areas (Park and Rukumnuaykit, 2004) and in rural migrant families than in local urban families. In addition, rural
families are more likely to have an additional child if the first child is a girl (either legally with government approval
or illegally). Therefore, compared to other students, migrant girl students are likely to have more siblings (and these
siblings are more likely to be girls - Yamaguchi, 1989; Jensen, 2002) and thereby may face greater competition in
intra-household resource allocation. Therefore, it is possible that the combination of a competitive environment with
a reminder of their inferior identity increased L girls’ effort, offsetting the adverse impact of stereotype threat.
27
28
performance on economically incentivized tasks. We answer this question in the context of the role
of the household registration or hukou system in generating different social identities in urban
China. Specifically, we compare the economic performance under incentives of two groups of
students who share different hukou identities: rural migrant students who are lower ranked hukou
holders in urban Beijing and their local urban counterparts who are higher ranked hukou holders.
We find that making individuals’ hukou identity salient significantly reduces the relative
performance of rural migrant students, compared to their local urban counterparts, in the piece rate
regime. In addition, migrant students’ average ranking of experimental earnings significantly
declines relative to local urban students under piece rate. This impact of hukou salience, however,
is insignificant in the tournament regime.
Our results are comparable to Hoff and Pandey (2012) who find that publicly revealing young
male students’ caste significantly decreases the responsiveness of low castes to incentives in rural
India. While they focus on male students this study shows that low status male and female students
are affected in a similar way.
Our study broadens perspectives on the impact of the hukou system on the rural-urban divide
to its impact on income gaps in urban areas of China through a hitherto unexplored channel. The
evidence in this paper underlines the causal effect of administratively created social identity in
distorting individuals’ performance on incentivized tasks and potentially exacerbating existing
inequities in the distribution of gains from economic growth. It shows that even in an ethnically
homogeneous society like China, several-decades-long systemic policy of social exclusion can
play a powerful role in creating identities and in turn influence individual behavior.
As suggested by Akerlof and Kranton’s (2000) identity model of social exclusion, the
negative impact of social exclusion is unlikely to disappear unless citizens are fully integrated into
29
a community. Future research should, thus, study the long-term impact of identity on educational
attainment and labor market outcomes of low status individuals. This strand of research could
provide critical insights for designing redistributive and inclusive policies that aim to achieve more
cohesive communities.
30
Acknowledgements
We are grateful to Karla Hoff for generously sharing her insights and experiment materials. We
are especially indebted to the co-editor and three anonymous referees for very helpful comments
and suggestions. We would also like to thank Rachel Croson, Catherine Eckel, Albert Park,
Ragan Petrie, Marco Castillo and seminar participants at the UT-Dallas behavioral/experimental
reading group, the International ESA meetings (Caltech, 2008), the Chinese Economist Society
annual meeting (Nanning, China, 2009), George Mason University, Florida State University, East
Carolina University, Peking University, the Indian Statistical Institute (Delhi), and the Delhi
School of Economics for their comments. Li acknowledges the financial support from the
National Science Foundation through grant no. SES 0720936. Any remaining errors are our own.
References
Akerlof, G.A. and Kranton, R.E. (2000). ‘Economics and Identity’, Quarterly Journal of
Economics, vol. 115(3), pp. 715–753.
Aronson, J., Quinn D. and Spencer S. (1998). ‘Stereotype Threat and the Academic
Underperformance of Minorities and Women’, in (J. K. Swim and C. Stangor, eds.), Prejudice:
The Target’s Perspective. New York: Academic Press.
Bargh, J.A. (2006). ‘What Have We Been Priming All These Years? On the Development,
Mechanisms, and Ecology of Nonconscious Social Behavior’, European Journal of Social
Psychology, vol. 36(2), pp. 147-168.
Benabou, R. and Tirole, J. (2007). ‘Identity, Dignity and Taboos: Beliefs as Assets’, CEPR
Discussion Paper No. DP6123.
Benjamin, D.J., Choi, J.J. and Strickland, A.J. (2010). ‘Social Identity and Preferences’, American
Economic Review, vol. 100(4), pp. 1913-1928.
Benjamin, D.J., Choi, J.J. and Geoffrey, F. (2013). ‘Religious Identity and Economic Behavior’,
NBER Working Paper 15925, August 2013.
Berg, A., Ostry, J.D. and Zettelmeyer J. (2012). ‘What Makes Growth Sustained?’, Journal of
Development Economics, 98(2): 149–166.
Berg, A. and Ostry J.D. (2011). ‘Inequality and Unsustainable Growth: Two Sides of the Same
Coin?’, IMF Staff Discussion Note 11/08.
Chan, K.W. (1996). ‘Post-Mao China: A Two-Class Urban Society in the Making’, International
Journal of Urban and Regional Research, vol. 20(1), pp. 134-150.
Chan, K. W. and Zhang, L. (1999). ‘The Hukou System and Rural–Urban Migration in China:
Processes and Changes’, The China Quarterly, vol. 160, pp. 818–855.
Charness, G., Rigotti, L. and Rustichini, A. (2007). ‘Individual behavior and group membership’,
American Economic Review, vol. 97, pp. 1340 – 1352.
Chen, A. (2002). ‘Urbanization and Disparities in China: Challenges of Growth and Development’,
China Economic Review, vol. 13(4), pp. 407–411.
Chen, R. and Chen, Y. (2011). ‘The Potential of Social Identity for Equilibrium Selection’,
American Economic Review, vol. 101(6), pp. 2562-2589.
Chen, Y. and Li, S. (2009). ‘Group Identity and Social Preferences’, American Economic Review,
vol. 99(1), pp. 431-457.
Cheng, T. and Selden, M. (1994). ‘The Origins and Social Consequences of China’s Hukou
System’, The China Quarterly, vol. 139, pp. 644–668.
31
Croizet, J. and Claire, T. (1998). ‘Extending the Concept of Stereotype Threat to Social Class: The
Intellectual Underperformance of Students from Low Socioeconomic Backgrounds’,
Personality and Social Psychology Bulletin, vol. 24(6), pp. 588-594.
Cohn A., Fehr E., Herrmann B., and Schneider F. (2011). ‘Social Comparison in the Workplace:
Evidence from a Field Experiment,’ IZA, Working Paper 5550.
Datt, G, and Ravallion, M. (2002). ‘Is India’s Economic Growth Leaving the Poor Behind?’,
Journal of Economic Perspectives, vol. 16(3), pp. 89-108.
Deaton, A. and Dreze, J. (2002). ‘Poverty and Inequality in India: A Re-examination’, Economic
and Political Weekly, vol. 37, pp. 3729-3748.
deBrauw, A. and Giles, J. (2008). ‘Migrant Opportunity and the Educational Attainment of Youth
in Rural China,’ World Bank, Working Paper 4526.
Dee, T. (2014). ‘Stereotype Threat and the Student-Athlete’, Economic Inquiry, vol. 52(1), pp.
173-182.
Eckel, C.C. and Grossman, P.J. (2005). ‘Managing Diversity by Creating Team Identity’, Journal
of Economic Behavior & Organization, vol. 58 (3), pp. 371–392.
Fang, H. and Loury G. (2005). ‘‘Dysfunctional Identities’ Can Be Rational’, American Economic
Review, vol. 95(2), pp. 104-111.
Fryer, R. G., Levitt, S. D. and List, J. A. (2008). ‘Exploring the Impact of Financial Incentives on
Stereotype Threat: Evidence from a Pilot Study’, American Economic Review, vol. 98(2), pp.
370–375.
Gneezy, U., Niederle, M. and Rustichini, A. (2003). ‘Performance in Competitive Environments:
Gender Differences’, Quarterly Journal of Economics, vol. 118(3), pp. 1049–1074.
Gneezy, U. and Rustichini, A. (2004). ‘Gender and Competition at a Young Age’, American
Economic Review, vol. 94(2), pp. 377–381.
Goette, L., Huffman, D. and Meier, S. (2006). ‘The Impact of Group Membership on Cooperation
and Norm Enforcement: Evidence Using Random Assignment to Real Social Groups’,
American Economic Review, vol. 96 (2), pp. 212–216.
Harrison, G.W., and List, J.A. (2004). ‘Field Experiments’, Journal of Economic Literature, vol.
42(4), pp. 1009-1055.
Hoff, K. and Pandey, P. (2006). ‘Discrimination, Social Identity, and Durable Inequalities’,
American Economic Review Papers & Proceedings, vol. 96(2), pp. 206-211.
_______ (2012). ‘Making Up People: The Effect of Identity on Preferences and Performance in a
Modernizing Society’, forthcoming, Journal of Development Economics.
Hoff, K. and Stiglitz, J. (2010). ‘Equilibrium Fictions: A Cognitive Approach to Societal Rigidity’,
American Economic Review Papers & Proceedings, vol. 100(2), pp. 141-146.
Horst, U., Kirman, A. and Teschl, M. (2006). ‘Changing Identity: The Emergence of Social
Groups’, Working Paper, University of British Columbia.
Jamieson, J. P., and Harkins, S. G. (2007). ‘Mere Effort and Stereotype Threat Performance
Effects’, Journal of Personality and Social Psychology, vol. 93, pp. 544-564.
Jensen, Robert (2002). ‘Fertility Preferences and Female Disadvantage: Equal Treatment,
Unequal Outcomes?’ mimeo, John F. Kennedy School of Government, Harvard University,
Cambridge.
Kanbur, R. and Zhang, X. (1999). ‘Which Regional Inequality? The Evolution of Rural–Urban
and Inland–Coastal Inequality in China from 1983 to 1995’, Journal of Comparative
Economics, vol. 27(4), pp. 686–701.
Lee, L. and Park, A. (2011). ‘Parental Migration and Child Development in China,’ mimeo, Hong
32
Kong University of Science and Technology.
Lin, J.Y, Cai, F., and Li, Z. (1996). The China Miracle: Development Strategy and Economic
Reform, Hong Kong, China: The Chinese University Press.
Liu, X. Y. (2011), Teaching Children Money Management, Tianjin Social Science Press, China
Tianjin.
Liu, Z. (2005). ‘Institution and Inequality: The Hukou System in China’, Journal of Comparative
Economics, vol. 33(1), pp. 133-157.
Lu, Z. and Song S. (2006). ‘Rural–Urban Migration and Wage Determination: The Case of Tianjin,
China’, China Economic Review, vol. 17(3), pp. 337–345.
Meng, X. and Yamauchi, C. (2012). ‘Impact of Rural-Urban Migration on Human Capital
Development of Migrant Children, ’ Paper presented at the 2012 PAA Meetings.
National Bureau of Statistics of China. 1994-2003. China Statistical Yearbook. Beijing, China:
China Statistical Press.
National Bureau of Statistics of China. (2002). Tabulation on the 2000 Population Census of the
People’s Republic of China, Beijing, China: China Statistical Press.
Niederle, M. and Vesterlund, L. (2007). ‘Do Women Shy Away from Competition? Do Men
Compete Too Much?’ Quarterly Journal of Economics, vol. 122(3), pp. 1067-1101.
Oswald, D. L., and Harvey, R. D. (2000). ‘Hostile Environments, Stereotype Threat, and Math
Performance among Undergraduate Women’, Current Psychology, vol. 19(4), pp. 338-356.
Park, A. and Rukumnuaykit, P. (2004). ‘Eat Drink Man Woman: Testing for Gender Bias in China
Using Individual Nutrient Intake Data’, Working paper, Oxford University.
Qu, Zhaopeng and Zhao, Z. (2012). ‘Evolution of the Chinese Rural-Urban Migrant Labor Market
from 2002 to 2007,’ mimeo, Renmin University of China.
Ravallion, M. and Chen, S. (2007). ‘China's (uneven) Progress against Poverty’, Journal of
Development Economics, vol. 82(1), pp. 1-42.
Shayo, M. (2005). ‘Nation, Class and Redistribution: Applying Social Identity Research to
Political Economy’, Working Paper, Princeton University.
Shih, M., Pittinsky, T.L. and Ambady, N. (1999). ‘Stereotype Susceptibility: Identity Salience and
Shifts in Quantitative Performance’, Psychological Science, vol. 10(1), pp. 81-84.
Solinger, D.J. (1999). Contesting Citizenship: Peasant Migrants, the State and the Logic of the
Market in Urban China. Berkeley, CA: University of California Press.
Steele, C.M. and Aronson, J. (1995). ‘Stereotype Threat and the Intellectual Test Performance of
African-Americans’, Journal of Personality and Social Psychology, vol. 69(5), pp. 797-811.
Wang, F. and Zuo, X. (1999) ‘Inside China’s Cities: Institutional Barriers and Opportunities for
Urban Migrants’, American Economic Review, vol. 89, pp. 276-280.
Whalley, J. and Zhang, S. (2007). ‘A Numerical Simulation Analysis of (Hukou) Labor Mobility
Restrictions in China’, Journal of Development Economics, vol. 83(2), pp. 392-410.
Xinhua News Agency. 2002. http://www.china.org.cn/english/China/36594.htm.
Yamaguchi, Kazuo (1989). ‘A Formal Theory for Male-Preferring Stopping Rules of
Childbearing: Sex Differences in Birth Order and the Number of Siblings,’ Demography,
26(3), p.451-65.
Yao, S., Zhang, Z. and Hanmer, L. (2004). ‘Growing Inequality and Poverty in China’, China
Economic Review, vol. 15(2), pp. 145– 163.
33
Table 1: Experimental Design
Treatments
1. Pure Piece Rate (PP)
control treatment
2. Mixed Tournament (PT)
control treatment
3. Pure Piece Rate (PP)
hukou salience treatment
4. Mixed Tournament (PT)
hukou salience treatment
Primed
hukou?
Reward per maze
Number of
sessions
Number of full
sessions*
Round
1
Round 2
Male
Female
Male
Female
No
¥1
¥1
9
6
7
5
No
¥1
¥6; winner takes all
9
5
7
5
Yes
¥1
¥1
15
7
14
6
Yes
¥1
¥6; winner takes all
15
6
12
5
Note: *The experiment consisted of 418 subjects in a total of 72 sessions, and 366 subjects in 61 full sessions (six subjects in
each session). There were 11 sessions with fewer than six subjects since some of the randomly selected students happened to be
absent from school on the day of the experiment.
34
Table 2: Summary Statistics of Pre-Experiment Survey Questions
Row No.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Variables [Survey Question No.]
Born in Beijing [Q1]
Years lived in Beijing [Q3]
Speak Beijing dialect at home [Q4]
Consider yourself as Beijing local [Q5]
Classmates consider you as Beijing local [Q6]
Teachers consider you as Beijing local [Q7]
Charged rural guest student fee [Q8]
Local Beijing students are more hardworking than
rural migrant students[Q10a]
Rural migrant students are more hardworking than
local Beijing students [Q10a]
Difference (row 9-8)
p value (row 9-8)
Local Beijing students are more active participants
in class than rural migrant students [Q10b]
Rural migrant students are more active participants
in class than local Beijing students [Q10b]
Difference (row11 – 10)
p value (row11 – 10)
Local Beijing students participate more in
extracurricular activities than rural migrant students
[Q10c]
Rural migrant students participate more in
extracurricular activities than local Beijing students
[Q10c]
Difference (row 13-12)
p value (row 13-12)
Local Beijing students are more careful with money
than rural migrant students [Q10d]
Rural migrant students are more careful with money
than local Beijing students [Q10d]
Difference (row 15-14)
p value (row 15-14)
L
H
0.202 (0.030)
5.713 (0.255)
0.716 (0.043)
0.093 (0.028)
0.231 (0.041)
0.236 (0.041)
0.600 (0.048)
0.164 (0.035)
0.705 (0.034)
7.342 (0.666)
0.946 (0.022)
0.864 (0.033)
0.817 (0.037)
0.900 (0.029)
0.036 (0.018)
0.135 (0.032)
Difference
(P-value)
0.000
0.006
0.000
0.000
0.000
0.000
0.000
0.552
0.327 (0.045)
0.369 (0.046)
0.511
0.164 (0.058)
0.005
0.245 (0.041)
0.234 (0.058)
0.000
0.414 (0.047)
0.440
0.136 (0.033)
0.171 (0.036)
0.473
-0.109 (0.053)
0.040
0.391 (0.047)
-0.243 (0.061)
0.000
0.486 (0.048)
0.140
0.145 (0.034)
0.144 (0.033)
0.978
-0.245 (0.060)
0.000
0.101 (0.029)
-0.342 (0.062)
0.000
0.126 (0.032)
0.309
0.633 (0.046)
0.595 (0.047)
0.558
0.532 (0.065)
0.000
0.468 (0.064)
0.000
0.497
0.008
0.152
0.556
Note: This pre-experiment survey, as part of the experimental intervention, was conducted at the beginning of the
salience treatment. Each row is cross-referenced in square brackets to a specific survey question in Appendix C. All
variables are reported as proportions (except row 2). Rows (3)-(6) are coded ‘1’ if the response to the question was 1,
otherwise ‘0’. Row (7) is coded ‘1’ if the categorical response was ‘1’, otherwise ‘0’. Rows (8) – (15) are coded ‘1’ if
the response to the question is as defined by the variable, otherwise ‘0’. For example, row (8) is coded ‘1’ if the
response is 1, i.e., “[l]ocal Beijing students are more hardworking,” to Q10a. Two-sided t test of differences in means
is used for row (2), and two-sided tests of proportions are used for all other comparisons. Missing responses are
excluded throughout. Standard errors are in parentheses.
35
Table 3: Summary Statistics
Variables
Grade
Age (years)
Born in Beijing
Years of local residence a
Father has college/graduate degree
Mother has college/graduate degree
Father in public/private enterprise
Mother in public/private enterprise
Played similar/same games before
Number of other subjects known
Pocket money given by parents
Control
N=72
(1)
4.389
(0.124)
9.778
(0.147)
0.625
(0.057)
8.063
(1.082)
0.422
(0.062)
0.381
(0.061)
0.714
(0.054)
0.667
(0.057)
0.653
(0.056)
1.338
(0.162)
2.859
(0.196)
H
Salience
Difference
Treatment (P value)
N=111
(2)
(1) vs. (2)
4.495
0.507
(0.101)
9.856
0.663
(0.108)
0.757
0.056
(0.041)
6.818
0.364
(0.844)
0.336
0.259
(0.045)
0.373
0.914
(0.046)
0.764
0.459
(0.041)
0.727
0.387
(0.042)
0.748
0.167
(0.041)
1.468
0.492
(0.111)
2.418
0.069
(0.147)
Control
N=72
(3)
4.167
(0.122)
9.944
(0.169)
0.208
(0.048)
5.617
(0.438)
0.074
(0.032)
0.087
(0.034)
0.514
(0.060)
0.507
(0.060)
0.750
(0.051)
1.586
(0.188)
2.443
(0.197)
L
Salience
Treatment
N=111
(4)
4.171
(0.104)
9.658
(0.127)
0.198
(0.038)
5.780
(0.309)
0.147
(0.034)
0.120
(0.031)
0.559
(0.047)
0.577
(0.047)
0.613
(0.046)
1.252
(0.129)
2.270
(0.145)
Difference
(P value)
(3) vs. (4)
0.978
Notes: Two-sided t tests of differences in means are used for grade, age, years of local residence in Beijing, number of other
subjects known and pocket money given by parents. Two-sided tests of proportions are used for all other variables. See coding
details for all variables in Appendix F. Standard errors are reported in parentheses. We run the F-test of the hypothesis that
all variables are jointly zero by regressing the salience treatment dummy on all the listed variables here, including a dummy
for missing values. The P-value of the F test of the joint significance of these variables is 0.316 for H and 0.298 for L subjects.
36
0.170
0.868
0.756
0.143
0.484
0.561
0.363
0.054
0.132
0.474
Table 4: Effects of the Hukou Salience Treatment on Performance (OLS)
Sample
Payment regime
Pooled
Piece rate &
tournament
Gender group
All
(1)
L (𝛽1 )
0.323
(0.364)
salience treatment (𝛽2 )
0.167
(0.359)
L*salience treatment (𝛽3 )
-0.880*
(0.466)
piece rate round 2
2.285***
(0.167)
tournament
2.508***
(0.173)
male
1.186***
(0.250)
grade
0.853***
(0.101)
played before
-0.304
(0.245)
Constant
1.956***
(0.604)
Observations
732
R-square
0.335
(a) Treatment effect
-0.714**
on L (𝛽2 + 𝛽3 )
(0.301)
Test of equality of coefficients (P values):
(b) Treatment effect on H (𝛽2 )
(c) Treatment effect on L (𝛽2 + 𝛽3)
(d) Difference in difference (𝛽3 )
One-way split
Piece rate
Tournament
All
(2)
0.583
(0.386)
0.329
(0.374)
-1.241**
(0.490)
2.266***
(0.167)
All
(3)
-0.466
(0.588)
-0.258
(0.599)
0.144
(0.800)
Two-way split
Piece rate & tournament
Boys
(4)
0.0779
(0.506)
-0.142
(0.477)
-0.615
(0.622)
2.447***
(0.213)
2.594***
(0.208)
Piece rate
Tournament
Girls
(5)
0.712
(0.493)
0.642
(0.538)
-1.208*
(0.696)
1.980***
(0.267)
2.339***
(0.308)
Boys
(6)
0.445
(0.526)
0.139
(0.498)
-0.963
(0.649)
2.427***
(0.213)
Girls
(7)
0.822
(0.532)
0.643
(0.564)
-1.686**
(0.740)
1.965***
(0.266)
Boys
(8)
-1.131
(0.790)
-0.838
(0.730)
0.374
(1.025)
Girls
(9)
0.477
(0.875)
0.502
(1.066)
0.278
(1.332)
1.179***
1.240***
(0.272)
(0.422)
0.766***
1.131***
(0.108)
(0.170)
-0.0543
-1.059**
(0.251)
(0.438)
2.002***
4.122***
(0.677)
(0.962)
558
174
0.300
0.284
-0.913***
-0.114
(0.323)
(0.508)
(2) vs. (3)
0.320
0.970***
0.612***
(0.128)
(0.160)
-0.172
-0.548
(0.313)
(0.376)
2.709***
2.886***
(0.697)
(0.877)
480
252
0.329
0.273
-0.757*
-0.566
(0.404)
(0.429)
(4) vs. (5)
0.275
0.897***
(0.135)
0.0907
(0.320)
2.546***
(0.771)
366
0.290
-0.823*
(0.426)
(6) vs. (7)
0.502
0.515***
(0.173)
-0.322
(0.389)
3.131***
(0.960)
192
0.235
-1.043**
(0.476)
(6) vs. (8)
0.173
1.207***
(0.212)
-1.037*
(0.541)
5.657***
(1.064)
114
0.300
-0.464
(0.681)
(8) vs. (9)
0.296
0.898***
(0.279)
-1.122
(0.749)
4.618***
(1.350)
60
0.235
0.780
(0.735)
(7) vs. (9)
0.892
0.130
0.746
0.730
0.599
0.214
0.025
0.084
0.524
0.461
0.184
0.954
0.143
Notes: School fixed effects are included throughout. Standard errors clustered at the individual level are in parentheses. * p < 10%, ** p < 5%, and *** p < 1%.
37
Table 5: Effects of the Hukou Salience Treatment on Experimental Earnings Distribution (OLS)
Sample
Payment regime
Pooled
Piece rate &
tournament
Gender group
All
(1)
L (𝛽1 )
1.500
(4.303)
salience treatment (𝛽2 )
2.546
(4.255)
L*salience treatment (𝛽3 )
-8.911
(5.479)
piece rate round 2
2.200
(1.715)
tournament
-28.51***
(2.436)
male
11.03***
(2.959)
grade
8.759***
(1.161)
played before
-2.262
(2.833)
Constant
-0.249
(7.083)
Observations
732
R-square
0.258
(a) Treatment effect
-6.364*
on L (𝛽2 + 𝛽3 )
(3.420)
Test of equality of coefficients (P values):
One-way split
Piece rate
Tournament
All
(2)
6.084
(4.456)
4.258
(4.269)
-13.500**
(5.660)
2.149
(1.708)
All
(3)
-12.32
(7.780)
-2.063
(8.223)
4.096
(10.61)
Two-way split
Piece rate & tournament
Boys
(4)
2.784
(5.908)
1.912
(5.532)
-7.799
(7.312)
2.781
(2.165)
-32.98***
(3.011)
Girls
(5)
0.0812
(6.111)
4.407
(6.671)
-12.24
(8.197)
1.156
(2.822)
-20.09***
(4.034)
Piece rate
Tournament
Boys
(6)
4.477
(6.116)
1.955
(5.751)
-9.931
(7.531)
2.784
(2.165)
Girls
(7)
8.816
(6.144)
8.176
(6.326)
-19.52**
(8.502)
1.020
(2.796)
Boys
(8)
-2.906
(10.67)
2.115
(10.33)
-1.722
(14.00)
Girls
(9)
-25.59**
(11.01)
-7.570
(14.01)
8.825
(15.72)
10.32***
(1.556)
1.254
(3.620)
2.052
(8.934)
366
0.154
-7.976
(4.931)
(6) vs. (7)
6.222***
(2.094)
-3.760
(4.709)
7.904
(11.26)
192
0.107
-11.349**
(5.614)
(6) vs. (8)
7.497**
(3.166)
-7.305
(7.244)
-13.48
(17.08)
114
0.065
0.393
(8.940)
(8) vs. (9)
10.56***
(3.607)
-7.699
(8.875)
-6.133
(18.66)
60
0.265
1.254
(6.044)
(7) vs. (9)
14.05***
1.445
(3.138)
(5.581)
8.918***
8.311***
(1.255)
(2.378)
-0.457
-7.942
(2.883)
(5.776)
-5.505
-12.73
(7.791)
(13.69)
558
174
0.179
0.110
-9.242**
2.033
(3.766)
(6.237)
(2) vs. (3)
9.618***
7.365***
(1.484)
(1.867)
-0.689
-5.069
(3.674)
(4.384)
6.249
8.435
(8.269)
(10.70)
480
252
0.272
0.189
-5.886
-7.832*
(4.754)
(4.525)
(4) vs. (5)
(b) Treatment effect on H (𝛽2 )
0.433
0.773
0.466
0.988
0.575
0.229
(c) Treatment effect on L (𝛽2 + 𝛽3 )
(d) Difference in difference (𝛽3 )
0.097
0.767
0.651
0.358
0.937
0.143
0.101
0.686
0.398
0.555
0.615
0.076
Notes: The dependent variable is individual’s earning percentile in the corresponding experimental round. Standard errors clustered at the individual level are in
parentheses. School fixed effects are included throughout. * p < 10%, ** p < 5%, and *** p < 1%.
38
Figure 1: Performance by Treatment and Hukou Type
9
8
7
6
5
9
8
H
L
8
7
5
L
5
H
6
5
6
7
8
Mazed solved (mean)
9
10
Mixed Tournament Regime Girls Only
9
10
Piece Rate Regime Girls Only
Mazed solved (mean)
10
9
8
L
6
7
5
H
Girls Only
L
Mixed Tournament Regime Boys Only
7
8
7
6
5
L
H
10
L
Mazed solved (mean)
10
H
Piece Rate Regime Boys Only
9
Mazed solved (mean)
10
9
8
7
6
H
Mixed Tournament Regime Only
6
L
Boys Only
5
Mazed solved (mean)
Mazed solved (mean)
10
8
7
6
5
H
Mazed solved (mean)
Piece Rate Regime Only
9
Mazed solved (mean)
9
8
7
6
5
Mazed solved (mean)
10
Pooled
10
Number of mazes solved
H
L
H
Control
Note: Bars represent 95% confidence intervals.
39
Salience
L
Figure 2: CDFs of Experimental Earnings by Treatment and Hukou Type
Control
Salience
40